In this paper, a new algorithm for MRI Brain Segmentation is proposed, which is based on the AntPart algorithm [1]. This algorithm proposed partitiones the brain structure into three parts-white matter, grey matter, and cerebrospinal fluid according to the grayvalues of pixels. The main algorithm compares each pixel with the nearest class center C, all the data belonging to class C, all the data carried by one ant chosen randomly and all the data at(at most) four steps in order, and optimize the performance by recalculating the class centers, merging the close classes and reclassfying the data classified wrongly. The performances of this algorithm are compared with those of two other ones: K-means and FCM. The results show that the method proposed is more accurate.